Based On Retinex Low-light Image Enhancement

As the carrier of information,image is widely used in communication,traffic,monitoring,remote sensing,robotics and other fields.Among the massive images,low illumination images account for a large proportion due to the shooting environment.Low-illumination images generally have the disadvantages of low brightness,uneven illumination,damaged details,serious color distortion,and high noise,which greatly reduce the value of use.Therefore,it is of great practical significance to enhance low illumination image.Retinex algorithm has color constancy,and can simultaneously achieve brightness enhancement,detail enhancement and color fidelity,which improves the quality of low illumination image more comprehensive.In order to solve the problems of low contrast and invisible details of partial images under uneven illumination and darkness,this thesis proposed two image enhancement methods based on Retinex theory.One is image enhancement method based on multi-level fusion and detail recovery.Firstly,this thesis converted the input image into HSV space,and copied the V channel equivalently into three layers: Retinex enhancement layer,brightness enhancement layer and detail prominence layer.Secondly,in Retinex enhancement layer,weighted guided filtering and morphology are combined to eliminate halo phenomenon,and global brightness and local detail adjustment factors are introduced to improve the single-scale Retinex model to enhanced brightness and details of images;In the brightness enhancement layer,a new normalization function is proposed to further enhance the brightness of the image by using the properties of inverse trigonometric function;In detail prominence layer,a local linear model is optimized and improved by using artificial bee colony algorithm to enhance details.Then three layers are fused by Laplace.According to Gamma correction characteristics and neighborhood pixel relations,a detail recovery scheme is proposed to avoid some details blurred after fusion.Finally,the experimental simulation results show that the algorithm can effectively enhance the contrast and details of the image both subjectively and objectively.The other is improved multi-scale Retinex algorithm for image enhancement.This thesis converted the input image into HSV space,and copied the V channel equivalently into two layers: improved Retinex enhancement layer and detail recovery layer.In the Retinex enhancement layer,three different scales of Gauss filtering are applied to the V-channel.The average value of the filter output is used as the brightness estimation,which achieves the effect of taking into account details enhancement,contrast enhancement and color consistency.The single-scale Retinex model is improved by using global brightness adjusting factor and local detail adjusting factor to enhance details and brightness.In detail recovery layer,three different scales of Gaussian blurring are used for V channel to achieve the effect of detail recovery.The results of the two layers are fused in a multi-scale manner.Finally,the simulation results show that the proposed algorithm can effectively enhance the image contrast and has the minimum lightness order error.